Lost in translation? When talking to a chatbot, the better word is getting lost in conversation.
In conversations with an AI model, users can ask anything and expect everything. Context that builds progressively through the exchanges can dilute the conversation as it drifts far from the initial topic.
Manus wants to address this challenge.
And that is by making its AI build context to include system instructions, uploaded files, conversation history, and any outputs or actions completed so far.
Using what it calls the 'Branch' feature, the AI can then enable the creation of a parallel session from any point in that history.
The new session receives a complete copy of everything accumulated up to the chosen message.
According to the announcement, the inheritance preserves instructions, attached files, and the full record of prior exchanges. The original session remains untouched and can continue independently. No changes or outputs from the branch affect the source conversation.
To start a branch, a user hovers over any earlier message in the chat interface.
A branch icon appears beneath the message.
Selecting it triggers the creation process. Confirmation appears quickly, after which the new session opens with the inherited materials ready for use. A breadcrumb indicator links the branch back to its origin for easy reference.
Isolation ensures that each branch maintains its own separate state after the divergence point.
Work in one branch does not interfere with others or with the original session.
Multiple branches can stem from the same message.
Branches can also extend from points inside other branches.
This supports tree like structures where different paths explore alternatives while sharing the same foundational context.
One application occurs in research workflows. A main session gathers data and assembles initial findings.
Once the core information exists, branches can generate distinct deliverables. One branch formats an investor brief that highlights financial projections and risks. Another branch produces an internal analysis report focused on operational details. A third branch derives action items and timelines.
The source session keeps the raw notes and data collection intact for further reference or continuation.
In content development, a session might complete a full draft along with supporting research.
A branch from that point adapts the material for another format, such as video scripting. The branch inherits the narrative structure, facts, and any reference files. It then applies adjustments specific to the new format, including pacing and visual cues.
The original draft session stays unchanged.
For ongoing projects, a persistent session functions as a central record.
It logs decisions, collected information, and progress over time. Branches created at milestones extract specific outputs, such as updated task lists or status summaries.
The central record continues to accumulate new entries without disruption from the derived work.
The feature reduces repeated effort in complex tasks where substantial context has already been established.
Context construction demands extended interaction to convey requirements, provide resources, and allow the agent to complete preliminary steps.
Branching reuses that investment across multiple objectives. Isolation maintains focus in each direction. Traceability through breadcrumbs allows users to review shared history or return to the source session as needed.
Currently the capability operates in standard chat sessions. It does not extend to Web Builder sessions. Future updates may broaden its availability.
Branch transforms session management from a strictly linear process into one that supports divergence at chosen points.
Accumulated understanding becomes reusable across parallel directions.
It avoids the overhead of restating background details or reattaching resources in new conversations. The result is more flexible handling of non linear work within the same agent environment.
